Analysing the Impact of MAUP

on the March of Atopy in England

using Hospital Admission Data

Nick Bearman

Nicholas J. Osborne & Clive Sabel

Associate Research Fellow in GIS

Introduction

Overview

• Terms & Literature Review– March of Atopy– MAUP– Hospital Episode Statistics

• Results• Conclusions &

Future Directions

The March of Atopy

“The atopic march ... describes the progression of atopic disorders, from eczema in young infants ... to allergic rhinitis and ... asthma in older ... children”Ker and Hartert (2009) p.282

Donell (2011)

Literature Review

March of Atopy

Barnetson (2002) Fig. 1

The March of Atopy is a contested area

Studies show that childhood eczema /

allergies significantly increases the

chance of asthma in later life

Literature Review

• Why is this link important?• Managing adult asthma costs the NHS

around £1bn per year (Gupta et al., 2004)

• If the March of Atopy is correct• Then reducing eczema will have a

massive long term cost saving

Literature Review

Issues with using Health Data

• Health data are aggregated into larger units, so individuals can’t be identified– GP Practice, PCT (Primary Care Trust)– SHA (Strategic Health Authority) – County, Post Codes, Census Output Areas,

Wards

• As users of GIS we know this matters…

Literature Review

MAUP – Modifiable Areal Unit Problem

An example....

A B

5 4

Literature Review

A second example

C D E

7 1 1

Literature Review

Same data – different message

C D E

7 1 1

A B

5 4

Literature Review

Literature Review

When point data are aggregated into

polygons,

any resulting summary data are influenced by

the choice of area boundaries C D E

7 1 1

A B

5 4 ≠

• March of Atopy is a contested concept • Many studies do not discuss aggregation of

data• Often epidemiological studies ignore MAUP

• Is MAUP partly responsible for the contested nature of the March of Atopy?– Compare the relationship between

asthma, eczema and allergy data– At different spatial aggregation levels

Literature Review

Methods

Hospital Episode Statistics (HES) data

• All admissions of patients to hospital – emergency (e.g. A&E) and referral (e.g. from

GP)

• Positives – Good spatial coverage for England– Comprehensive

• Negatives– Only capture severe cases– Issues with small numbers / confidentiality

Lightfoot

Methods

Method

• Using ICD-10 codes to define disease (Anandan et al., 2009)

• Filtered by PatientID, so per person• Calculate age-sex directly standardised rates

for– Eczema 0-14 years, Allergy 0-14 years– Asthma 15+ years

• 2008/9 to 2010/11 (3 years)• Data for England• PCT (152), LA (354)

Methods

Results

Asthma age-sex directly standardised rates per 1000 by LA

Results

• Impact of presence of Eczema on likelihood of suffering Asthma– Allergy not significant– Also corrected for

IMD, which made no difference

– No significant difference between PCT & LA

N Eczema ConstantCoef. p 95% Conf. Int. Coef. p

PCT 123 2.21 0.002 0.81 3.60 7.36 <0.001LA 168 1.13 0.001 0.45 1.82 6.86 <0.001

Linear Regression

Results

Limitations

• Ideally need eczema and asthma data for same individuals

• But these are difficult/expensive to access and don’t exist with sufficiently detailed medical reporting

• Assuming individuals don’t move around too much / rates in one place don’t change over time

Results

• Missing ~20% of cases at LA level because data <6 cases per sex per year are supressed

• Did subset comparison of PCT & LA complete data

Results

Results

Results

Clustering

• Can look for unusual clusters• Where values are higher (or lower) than

expected

• Use univariate LISA:

– To identify statistically significant clusters

Local Indicators of Spatial Autocorrelation

(Anselin, 1995)

LISA

Also highlights some of the loss of detail moving from LA to PCT

Next Stages

PCT -> LA -> GP?

• Next stage would be to look at data by GP (n~8000)

• But HES data where n < 6 is supressed• Unsupressed data costs too much to

access– (~£1500)

• (Hopefully) be able to get data by Postcode District (n~2000) from Met Office

Conclusions

Conclusions

• For this instance of the March of Atopy, using these data, MAUP is not an issue

• Need more data to further explore the issue– Ideally GP, but Postcode District will provide

more information

• Moving from LA to PCT can result in a loss of data

www.ecehh.orgn.bearman@exeter.ac.uk

Questions?

Regression with Allergy  N Eczema Allergy Constant    Coef. p 95% Conf. Int. Coef. p Coef. pPCT 123 2.19 0.003 0.74 3.64 0.05 0.923 7.33 <0.001LA 168 1.21 0.001 0.47 1.94 -0.20 0.545 7.02 <0.001

N Eczema ConstantCoef. p 95% Conf. Int. Coef. p

PCT 123 2.21 0.002 0.81 3.60 7.36 <0.001

LA 168 1.13 0.001 0.45 1.82 6.86 <0.001

Subset Analysis

Original Subset

Subset Analysis

Age-sex specific rates per 1000 population per year

PCT LA

Asthma (aged 15+)Mean 10.086 10.602St Dev 2.339 2.410

Eczema (aged 0-14)Mean 0.226 0.283St Dev 0.402 0.605

Allergies (aged 0-14)

Mean 0.961 0.913St Dev 0.346 0.340

  N Eczema Allergy Constant    Coef. p 95% Conf. Int. Coef. p Coef. p

PCT 84 1.43 0.028 0.16 2.71 0.57 0.441 9.21 <0.001LA 87 0.93 0.035 0.07 1.79 0.78 0.318 9.63 <0.001

  N Eczema Constant    Coef. p 95% Conf. Int. Coef. p

PCT 84 1.55 0.014 0.32 2.79 9.73 <0.001LA 87 1.06 0.013 0.23 1.89 10.30 <0.001

MAUP – Scale and Aggregation/Zoning Effects

From: http://www.geog.ubc.ca/courses/geog570/talks_2001/scale_maup.html